Bachelor’s Degree A - Home | Scholars at Harvard · Bachelor’s Degree Attainment and Time to...

Aiding or Dissuading? The Effects of Exhausting Eligibility for Need-based Aid on

Bachelor’s Degree Attainment and Time to Completion

Zachary Mabel

Harvard Graduate School of Education

August 2017

ABSTRACT

While studies find that need-based financial aid can increase college access,

persistence, and completion, little is known about how tuition subsidies affect

students after they have enrolled in college and how losing aid impacts

postsecondary attainment. I shed light on these questions by exploiting recent

changes to federal Pell Grant eligibility rules that reduced the lifetime availability

for aid from 9 to 6 full-time-equivalent years. Using longitudinal data from the

University System of Georgia, I compare student outcomes before versus after the

rule change for Pell recipients who were affected and unaffected by the new

lifetime limit. The results indicate that aid exhaustion required students to borrow

more, which in turn accelerated their time to completion. The rule change

increased the probability of term-over-term re-enrollment and bachelor’s degree

completion within 8 years by 2-3 percentage points. Students who had at least one

year to adjust to the new rule before exhausting aid eligibility were also much

more likely to graduate before experiencing aid losses. Importantly, I find no

evidence that the rule change decreased the overall probability of degree

attainment. The findings therefore indicate that setting limits on the availability of

need-based grants can accelerate time to completion for students who have

demonstrated a sustained commitment to finishing college.

I am grateful to the Office of Research and Policy Analysis at the Board of Regents of the University

System of Georgia, and in particular to Angela Bell, Rachana Bhatt, and Ke Du, for making this study

possible by sharing their data. Thank you also to Christopher Avery, Celeste Carruthers, David Deming,

Bridget Terry Long, Martin West and participants at the EPPE Colloquium at the Harvard Graduate School

of Education, APPAM Fall 2016 conference, and SREE Spring 2016 conference for helpful comments and

suggestions on earlier versions of this paper. All errors, omissions, and conclusions are mine.

1

I. INTRODUCTION

For decades financial aid has been a widely utilized strategy to support access to higher

education and postsecondary attainment. In the fifty years since the passage of the federal Higher

Education Act of 1965, average aid per student has more than tripled in real dollars, from $3,800

(in 2017 dollars) to $14,500, largely due to the expansion of federal aid programs (Baum, Elliott,

& Ma, 2014; Dynarski & Scott-Clayton, 2013). Despite this growth in spending, many students

who attend college withdraw before earning a certificate or degree. Less than one-third of

degree-seeking students who enter community colleges earn an associate’s or bachelor’s degree

within six years of initial enrollment, and nearly 40 percent of students who begin at four-year

institutions exit without a degree.1 The size of aid expenditures and magnitude of dropout have

motivated questions about whether financial aid is effectively helping students progress to

graduation.

While causal research finds that generous and simple aid programs can increase access

and produce long-term impacts on completion and earnings (Angrist, Autor, Hudson, & Pallais,

2016; Bettinger, Gurantz, Kawano, & Sacerdote, 2016; Castleman & Long, 2016; Dynarski,

2003; Goldrick-Rab, Kelchen, Harris, & Benson, 2016), little is known about how financial aid

affects students after they have enrolled in college. In particular, although most graduates take

longer than is customary to finish, it is unclear whether aid that gets disbursed beyond the second

year of college affects whether students will graduate and how quickly they do so.2

1 Authors’ calculations using the U.S. Department of Education’s National Center for Education Statistics

2004/2009 Beginning Postsecondary Students Survey. 2 The majority of students do not graduate on time for several reasons: 1) approximately 35 to 40 percent of them are

required to take developmental education courses before they can progress towards degree requirements (Bettinger,

Boatman, & Long, 2013); 2) most students also work while attending school, hindering full-time enrollment and

slowing progress to completion (Bound, Lovenheim, & Turner, 2012; Scott-Clayton, 2012); and 3) discontinuous

enrollment is also widespread. Using administrative data from Florida and Ohio, I calculate that nearly one-third of

students take time off for at least one semester before returning to pursue their degree.

2

Millions of college students also lose financial aid each year (Schudde & Scott-Clayton,

2016; Suggs, 2016), yet evidence on whether losing aid affects attainment is limited and

inconclusive. Carruthers & Özek (2016) find that losing merit-based aid in Tennessee

accelerated when students left college but had no effect on overall degree completion, while

Schudde & Scott-Clayton (2016) find evidence that failure to meet academic requirements for

Pell renewal increased dropout in the short-run but also potentially increased long-run transfer

and completion for students who persisted. Furthermore, because the studies above have only

examined effects of aid loss early in college due to poor academic performance, the findings may

not generalize to contexts in which aid is taken away later in college and for non-academic

reasons.3 Knowing whether the effects of aid vary over time and how students respond when aid

is taken away thus has important implications for maximizing the allocation of aid dollars to

support college student success.

I shed light on these questions by examining the impact of exhausting eligibility for need-

based aid on long-term student outcomes. Using matched difference-in-differences (DD) and

difference-in-difference-in-differences (DDD) designs, I exploit recent changes to federal Pell

Grant eligibility rules that reduced the lifetime cap on aid from 9 to 6 full-time-equivalent (FTE)

years beginning in the 2012-13 school year. Because it immediately eliminated a subset of

continuing students from receiving need-based aid and reduced award amounts for others, the

rule change provides a source of plausibly exogenous variation to estimate the effects of

exhausting Pell Grant eligibility on several outcomes, including how much students borrow to

pay for college, their investments in academic effort, and the probability of bachelor’s degree

attainment and time to completion.

3 For example, in their study of the effects of post-9/11 price shocks on undocumented students in New York City,

Conger & Turner (2015) find that tuition increases affected the likelihood of completion for recent entrants but not

for students who had enrolled for two or more years when costs increased.

3

To preview my results, I find that students lost one-third of their grant aid after the new

lifetime limit took effect. Students compensated for this loss by borrowing 15 percent more per

year to pay for college and increasing their academic effort. The eligibility change increased the

probability of term-over-term re-enrollment and bachelor’s degree completion within 8 years by

2-3 percentage points, and students who had at least one year to adjust to the new rule before

exhausting their eligibility were much more likely to graduate before experiencing any loss of

aid. Importantly, I find no evidence that the rule change decreased the probability of degree

completion overall. These findings indicate need-based aid has heterogeneous effects along the

path to graduation. For students who have demonstrated a sustained commitment to finishing

college, setting limits on the availability of need-based aid can accelerate time to completion.

I structure the remainder of the paper as follows. In Section II, I discuss the theoretical

motivation for this study and provide details on the change to Pell Grant eligibility. I describe the

data, analytic samples, and research design in section III. In section IV, I present the empirical

results. I conclude in section V with a discussion of the findings and directions for future

research.

II. BACKGROUND

Competing hypotheses posit that financial aid may help or have little impact on progress

to degree completion. On the one hand, grant aid may persuade students on the margin of

graduating to persist by reducing out-of-pocket expenses. According to standard human capital

theory, students are expected to choose to enroll in an additional year of college if the expected

lifetime benefit of attending an extra year exceeds the expected lifetime benefit of dropping out.

Because this model assumes that students update their expectations with experience, decisions to

persist in college may be influenced by changes to the availability of financial aid. For students

4

on the margin of graduating, losing aid may alter the cost-benefit evaluation enough to induce

departure.

In addition to changing the investment value of attendance, students may become

acclimatized to receiving aid, making it difficult to forecast and contingency plan for abrupt

changes in funding. Older students may also face stiffer credit constraints than their younger,

financially dependent peers, making it more difficult to offset grant losses with more student

loans later in college (Gichevu, Ionescu, & Simpson, 2012). Even in the absence of borrowing

constraints, loan aversion, which is pervasive among college-goers, may dissuade students from

replacing grants with loans (Boatman, Evans, & Soliz, 2017; Goldrick-Rab & Kelchen, 2015).

All of these scenarios predict that losing eligibility for need-based aid will decrease the

probability of persistence towards the beginning and end of college. Furthermore, because

individuals weigh losses more than gains (Kahneman & Tversky, 1979), losing aid may in fact

have greater consequences on degree completion than receiving aid of equal value.

On the other hand, the effect of financial aid on attainment may diminish as progress is

made in school. For instance, students may become less likely to update their schooling decisions

over time as their investments in college accumulate and the remaining cost to completion

declines. Decisions to persist may stabilize over time for this reason and attenuate the impact of

aid on persistence as experience accrues. Furthermore, because grant aid offsets the full cost of

attendance to students, offering aid late into college may also delay graduation for some students.

Several studies find that students respond to the availability of financial aid by strategically

adjusting their enrollment intensity to meet renewal requirements (Cornwell, Lee, & Mustard,

2005; Richburg-Hayes et al., 2009; Scott-Clayton, 2011), and while most studies find that these

effects are tied to performance incentives, recent experimental evidence finds that the amount of

5

time grant aid is offered can delay time to completion even when few strings are attached

(Angrist et al., 2016). If time to completion is partly a function of the financial pressure students

feel to enter the labor market, then eliminating aid eligibility late into college could increase the

efficiency of degree production.4

Empirically, whether the duration that need-based aid is offered affects progress to

completion remains an open question. By focusing on initial or cumulative aid amounts, prior

studies have largely ignored whether enrollment duration moderates the effect of aid on

persistence and completion. DesJardins, Ahlburg, & McCall (2002) do find evidence that

frontloading grants and scholarships would lead to modest increases in persistence at four-year

college; however, the robustness and generalizability of this study is unclear because it is based

on simulated policy changes using a dated sample of students that enrolled at a single institution.

Extending the Literature by Examining Effects of Reducing Lifetime Pell Eligibility

In this study, I examine how exhausting eligibility for federal Pell Grant aid affects

borrowing and enrollment decisions for students who are close to graduating. As my review of

the literature reveals, no studies have isolated the effect of aid on persistence as it is disbursed

late into college. To the extent that aid increases postsecondary attainment, effects may be driven

by early subsidies that set students on a path they would follow in the absence of continued

support. Alternatively, financial constraints may pose a formidable barrier to attainment along

the entire pathway to completion. The findings in this paper therefore help to tease out whether

the effects of need-based aid vary with time spent in college and how students who have already

made considerable educational investments respond to losing grant dollars.

4 The decision to delay completion may also intensify in weak economic cycles when labor market opportunities for

recent graduates are less certain. The aftermath of the Great Recession, which coincided with enactment of the new

lifetime Pell limit, is one such period when financial aid might have provided a stronger inducement to forego

graduation until more promising job opportunities became available.

6

To identify effects on student outcomes, I exploit recent changes to Pell Grant eligibility

rules which took effect in the 2012-13 school year. In 2011, the Pell Grant program faced an $18

billion shortfall as a result of growing enrollments in college and recent program changes that

made more students eligible for aid.5 After infusing the program with $17 billion, Congress

addressed the remaining funding gap by implementing the following four eligibility changes

which applied to both incoming and continuing students:

1) Eliminating eligibility for students without a high school diploma or GED;

2) Eliminating eligibility for students who qualified for the smallest grant amount,

equivalent to 10 percent of the maximum award, or $555;

3) Reducing the family income ceiling that automatically qualified students for the

maximum award from $32,000 to $23,000; and

4) Reducing the lifetime duration of eligibility from 9 to 6 FTE years.6

I examine effects on student outcomes caused in particular by reducing lifetime eligibility for

Pell Grant aid.7 Estimates suggest that nearly 400,000 undergraduates were affected by this

provision alone and that students attending four-year institutions were disproportionately

impacted.8

5 The number of students receiving a Pell Grant increased by 13 and 27 percent in 2008-09 and 2009-10,

respectively, whereas the year-over-year increase never exceeded 5 percent between 2004-05 and 2007-08 (Mahan,

2011). While part of this increase is attributable to enrollment growth during the Great Recession, Congress also

relaxed income eligibility restrictions for Pell Grant aid and increased the maximum grant amount during this time,

both of which contributed to skyrocketing program costs (Alsalam, 2013). 6 As I describe in more detail in section III, the FTE provision means that lifetime Pell use is determined by two

factors: 1) how many years students have received Pell support, and 2) their enrollment intensity (i.e., full-time,

part-time, etc.) during those years. 7 All of the students in my analytic sample are high school graduates who qualify for Pell awards above the

minimum amounts. I examine empirically whether the income change for auto-zero qualification impacts student

outcomes. As I discuss in section IV, effects are concentrated among students who did not qualify for maximum

awards. I therefore find no evidence that the income restriction affected students’ schooling decisions. 8 For example, the California State University System predicts that 4 percent of its total undergraduate population

lost eligibility as a result of the new lifetime limit (Nelson, 2012). If this percentage is nationally representative, then

374,000 of the 9.35 million students enrolled in four-year degree programs are predicted to have been affected. I

find that four percent of students in the USG dataset were also potentially affected by the new lifetime limit.

However, the USG dataset excludes students who entered as transfer students, and as a result, the 400,000 estimate

may be conservative.

7

The U.S. Department of Education (DOE) first announced these eligibility changes to

institutions in January 2012, six months before they went into effect. However, DOE did not

contact students about the changes until April 2012, and the only students directly contacted (via

e-mail) were those who had received more than 4.5 FTE years of Pell (U.S. Department of

Education, 2012). Also in April, DOE provided institutions with a list of the students who were

e-mailed. Beginning in July 2012, students could check their lifetime Pell use by logging into the

National Student Loan Data System, and the federal aid processing system began automatically

flagging students in excess of 4.5 FTE years of Pell for school financial aid administrators. As a

result of this communication rollout, students and institutions had very little time to prepare for

the eligibility changes in 2012-13, while in later years more time and resources afforded students

greater opportunity to adjust their enrollment decisions.

III. EMPIRICAL ANALYSIS

Data

The data for this study are from the University System of Georgia Enterprise Data

Warehouse (USG), which maintains longitudinal student-level records at twenty-eight public,

four-year colleges and universities in the State of Georgia.9 The dataset includes records on

301,423 degree-seeking students who first attended a USG institution in the fall term between

2002 and 2008, inclusive. From this data I observe information on students at the time of

application, including their demographics and college entrance examination scores. The data also

contain the financial information that students and their families supplied when completing the

Free Application for Federal Student Aid (FAFSA) each year, including the Expected Family

Contribution (EFC) used to calculate how much federal Pell Grant aid students are eligible to

receive, and all financial aid disbursements students actually received while enrolled in college.

9 Since 2013, the Board of Regents of the University System of Georgia has consolidated eighteen institutions into

nine for cost-saving purposes. However, because the GDW data include institutional identifiers prior to

consolidation, the dataset in practice covers student enrollments across thirty-seven unique campuses.

8

Lastly, the dataset includes complete records of students’ enrollment, course-taking, and degrees

received across the USG system through summer 2016, as well as records of transfer into USG

from other postsecondary systems. Taken together, this rich dataset enables me to construct a

cumulative measure of lifetime Pell receipt for each student and examine how the threat or actual

loss of Pell Grant eligibility affects student borrowing and re-enrollment decisions, as well as the

probability of bachelor’s degree completion overall and within different time intervals.

Defining Treated Students

Because the dataset does not include a direct measure of cumulative Pell receipt, I

used the reported EFC and disbursed Pell Grant amounts to construct this measure for each

student and to identify those affected by the new lifetime rule. Specifically, I employed the

following algorithm to calculate the amount of lifetime Pell that students used over eight

years:

(1) 𝑇𝑜𝑡𝑎𝑙 𝑃𝑒𝑙𝑙 𝐹𝑇𝐸 𝑌𝑒𝑎𝑟𝑠𝑖 = ∑𝑃𝑒𝑙𝑙 𝑅𝑒𝑐𝑒𝑖𝑣𝑒𝑑𝑖𝑡

𝑀𝑎𝑥 𝑃𝑒𝑙𝑙 𝐸𝑙𝑖𝑔𝑖𝑏𝑙𝑒𝑖𝑡| 𝐸𝐹𝐶𝑖𝑡,𝐶𝑜𝑠𝑡𝑠𝑡 8𝑡=1 , where

the full-time-equivalent years of Pell received for student 𝑖 in year 𝑡 was calculated as the

amount of Pell aid received relative to the maximum amount the student was eligible to

receive to subsidize the cost of full-time enrollment at college 𝑠. To determine the maximum

eligible amounts, I relied on the annual Pell award disbursement schedules published by the

U.S. Department of Education (DOE), which identify the grant amounts students qualify for

as a function of their EFC and cost of attendance.10

In most years in this study, Pell recipients who enrolled full-time over an entire

school year received one FTE year of Pell. Recipients who enrolled less than full-time

received less than one FTE year of Pell, with the specific amount determined by the student’s

10

While the USG dataset includes the EFC for students who filed the FAFSA, it does not include the cost of

attendance that was charged to each student; nevertheless, this does not preclude using the disbursement schedules

to identify eligible award amounts because the cost to attend USG institutions is sufficiently high that eligible award

amounts in practice are based solely on EFC for the vast majority of students in the system. This holds, with very

few exceptions, for students across all school years, institutions, and living arrangements in this study.

9

EFC and enrollment intensity (i.e., whether the student enrolled three-quarters-time, half-

time, or less-than-half-time in each term that Pell aid was disbursed). An exception to these

rules occurred in the 2009-10 and 2010-11 school years, when Pell recipients could qualify

for a second award in the same year to subsidize the cost of summer attendance. In those two

years some students therefore accumulated more than one FTE year of Pell during a single

award year. Nationally, 1.2 million students (13 percent of all Pell Grant recipients) received

supplemental awards in 2010-11, which increased the average grant per recipient by

approximately $200, or 6 percent (Baum et al., 2014; Delisle & Miller, 2015). In my main

analytic sample, which I describe in the next section, 26 percent of students received more

than one FTE year of Pell in either 2009-10 or 2010-11.

While students remain eligible for Pell Grants until they receive 6 FTE years of Pell

under the new lifetime limit, I define treated students as those who enrolled on or after spring

2012 and received 5 or more FTE years of Pell (5 Pell FTE students). The rationale for

defining treated students more expansively than is set by the statutory limit is twofold. First,

under the new eligibility rules, students whose cumulative Pell receipt exceeds 5 FTE years

receive proportionally smaller award disbursements in new award years until they reach the 6

FTE year lifetime limit. For example, a student with an EFC of $1,000 who had received 5

FTE years of Pell through 2011-12 would have been eligible for a Pell Grant of $4,600 in

2012-13, whereas the largest grant available to a student with the same financial need that

year who had accumulated 5.5 FTE years of Pell would have been $2,300. Including 5 Pell

FTE students in the treated group therefore accounts for effects as a result of decreasing the

generosity of aid in addition to the effects from full exhaustion.

It is also possible that the new lifetime limit created anticipatory effects which altered

the probability of aid exhaustion in the post-policy period. Under the new rules, students may

have chosen to withdraw before they experienced award reductions if they knew they needed

to complete more than a year of coursework to graduate. Students aware of the new limit

10

could also have increased their enrollment intensity to graduate before experiencing aid

losses. As a result, including 5 Pell FTE students in the treated group also allows for

estimation of both direct and anticipatory policy effects.

Sample

If all Pell recipients made the same enrollment decisions over time, then students

affected by the new lifetime limit could be identified by when they first entered college. In

practice, which students are affected is more complicated because students’ enrollment

decisions are not uniform or random. Some students enroll continuously, while others stop

out. Some students also alter their enrollment intensity over time. Because these endogenous

decisions affect who is at risk of exhausting aid eligibility and they are correlated with the

probability of degree attainment, straightforward comparisons of treated and non-treated

students will lead to biased estimates of policy effects. A key challenge in this study is

therefore identifying an analytic sample that accounts for selection into the treatment group.

I address the selection issue by conditioning the sample in three steps. In step one, I

limited the sample to students for whom complete Pell receipt is observed by excluding

individuals who transferred into the USG system at any time. Of the 301,423 students in the

dataset, this restriction eliminated 18 percent of the sample. In step two, I further conditioned

on students who enrolled in college for 5 or more FTE years (5 FTE students) to isolate the

subset potentially impacted by the new lifetime Pell limit.11

46,766 students (19 percent of

the remaining sample) attained 5 FTE status. Lastly, in step 3, I restricted the dataset to a

matched sample of 5 FTE students in which non-5 Pell FTE students are observably similar

to the group of 5 Pell FTE students. Matching addresses the fact that even after conditioning

the sample on 5 FTE students, the majority of 5 Pell FTE students are observably different

11

I use a modified version of equation (1) to calculate FTE status for all students, where students’ EFC-eligibility

status is ignored and FTE status is determined by enrollment intensity during each academic year. For example, a

student who attempted 12 credits during fall 2010 and 6 credits during spring 2011 would be assigned an FTE of

0.75 for that year (i.e., 0.5 for full-time enrollment in the fall and 0.25 for part-time enrollment in the spring). A nice

feature of this approach is that Pell FTE years and FTE years are derived from a consistent set of rules.

11

from other 5 FTE students on key dimensions. Table 1 presents summary statistics by

treatment status for the overall 5 FTE sample and the subset of matched students. Compared

to the overall sample, 5 Pell FTE students are more likely to be female (62 percent vs. 56

percent) and Black (61 percent vs. 41 percent), have greater financial need according to their

EFCs at entry and in year 5 ($1,092 vs. $14,434 at entry, for example), and entered college

with lower average SAT scores (928 vs. 1,012).

I used the coarsened exact matching (CEM) procedure developed by Iacus, King, &

Porro (2012) to construct the matched sample. This procedure allows the researcher to

identify which characteristics to match on and specify how to coarsen the data for matching

(if at all), and then exactly matches treated and non-treated observations using the coarsened

data. CEM has a number of attractive properties over more traditional matching methods like

propensity score matching. First, it obviates the search for a suitable matching algorithm to

achieve ex-post balance. Second, it avoids creating uninformative matches that approximate

random matching which can produce more biased inferences than not matching at all (King,

Nielsen, Coberley, & Pope, 2011). CEM also ensures that matching on a subset of observed

variables has no effect on the imbalance of variables not used in the matching procedure.

Checking for balance on non-matched covariates therefore sheds light on the plausibility of

the key assumption of matching: that treatment is independent of potential outcomes

conditional on the covariates used to match treated and non-treated observations.

5 Pell FTE students were matched to other 5 FTE students using the following baseline

characteristics: entry cohort (not coarsened), sex, race (White, Black, or Other), an indicator of

whether the student enrolled continuously or stopped out in the first four years of college,

number of years to attain 5 FTE status (not coarsened), quartile of cumulative credits attempted

at the start of students’ 5 FTE year (not coarsened), EFC in the 5 FTE year ($0-$1,300, $1,300-

$2,600, $2,600-$5,200, > $5,200), and an indicator of whether each student enrolled on or after

12

fall 2012.12

This procedure generated a matched sample of 16,588 students composed of 8,656 5

Pell FTE and 7,932 other 5 FTE students.13

Characteristics of the matched sample are reported in columns 3-6 of Table 1. By design,

treated and comparison students exhibit the same mean characteristics on the variables that were

used in the matching procedure. For example, 62 percent of both groups are now female and the

mean number of credits attempted at the start of students’ 5 FTE year is 136 in both groups. In

addition, both groups are now quite similar on non-matched measures of academic performance

at entry and in college. The mean difference in SAT achievement between groups in the matched

sample is now just 22 points (928 vs. 950 among treated and non-treated students, respectively),

and students in both groups completed the same number of credits on average (121) and earned

the same average grades (2.65) at the start of their 5 FTE year.

The key difference that distinguishes 5 Pell FTE from other 5 FTE students is the number

of years of Pell received. On average, 5 Pell FTE students received 5.6 years of Pell compared to

3.2 years for other 5 FTE students. Because students were matched on their EFC in the 5 FTE

year to ensure treated and comparison students had similar income profiles at the end of college,

there are two major reasons why some students in the matched sample received less than 5 FTE

years of Pell. First, 21 percent of other 5 FTE students failed to file the FAFSA at the start of

college, whereas only 1 percent of 5 Pell FTE students did not submit the FAFSA in their first

year. Second, many other 5 FTE students who applied for aid were not initially eligible for a Pell

Grant. The average EFC at entry among other 5 FTE students was $6,642 (in 2016 dollars),

which exceeds the maximum value for Pell receipt. In summary, whereas 5 Pell FTE students

filed the FAFSA consistently and regularly received Pell aid, other 5 FTE students did not. As a

12

The upper bounds of the bottom three attempted credit quartiles are: 126, 136, and 146 credits. 13

88 percent of 5 Pell FTE students and 21.5 percent of other 5 FTE students are included in the matched sample.

13

result, their eligibility for need-based aid remained unaffected when Congress reduced the

lifetime limit to 6 FTE years.

My empirical strategy rests on comparing the outcomes of 5 Pell FTE students and other

5 FTE students in the matched sample before versus after the lifetime rule change took effect. To

obtain unbiased estimates of policy effects, I must therefore assume that differences between

students who enrolled before versus after the rule change are stable across groups in the matched

sample. If this holds, then I can obtain unbiased causal estimates by differencing out selection

effects observed within each group over time.

I examine evidence for differential selection in the matched sample in Table 2.

Columns 1-4 report group-specific means before versus after the rule change was introduced.

In columns 3 and 6, I present estimates of within-group selection effects. Column 7 shows

whether those changes differed for 5 Pell FTE and other 5 FTE students. The results in

column 3 indicate that 5 Pell FTE students who enrolled post-2011 entered college with more

financial resources than previously enrolled students, but were more academically

disadvantaged. However, because these differences (and all others reported in Table 2) are

also observed in the other 5 FTE group, there is no evidence of differential selection between

groups in the matched sample. Again, this holds by definition for the characteristics on which

students were matched as well as for all observable dimensions on which students were not

matched.

Empirical Strategy

I estimate intent-to-treat effects on students’ term-by-term enrollment decisions and

financial aid receipt using a matched difference-in-difference-in-differences (DDD) strategy,

where the first difference is before versus after the policy change, the second difference is

whether or not a student received 5 FTE years of Pell, and the third difference is whether or not

14

students had received more than 4.5 FTE years of Pell at the start of each academic term. This

design is implemented by fitting the following statistical model to a student-by-term dataset:

(2) 𝑌𝑖𝑗𝑡 = 𝛼1𝑇𝑟𝑒𝑎𝑡𝑖 ∗ 𝑃𝑜𝑠𝑡2011𝑡 ∗ 𝑃𝑜𝑠𝑡4.5𝐹𝑇𝐸𝑖𝑡 + 𝛼2𝑃𝑜𝑠𝑡2011𝑡 ∗ 𝑃𝑜𝑠𝑡4.5𝐹𝑇𝐸𝑖𝑡 +

𝛼3𝑇𝑟𝑒𝑎𝑡𝑖 ∗ 𝑃𝑜𝑠𝑡2011𝑡 + 𝛼4𝑇𝑟𝑒𝑎𝑡𝑖 ∗ 𝑃𝑜𝑠𝑡4.5𝐹𝑇𝐸𝑖𝑡 + 𝛼5𝑇𝑟𝑒𝑎𝑡𝑖 +

𝛼6𝑃𝑜𝑠𝑡2011𝑡 + 𝛼7𝑃𝑜𝑠𝑡4.5𝐹𝑇𝐸𝑖𝑡 + 𝜙 𝑋𝑖 + 𝜔𝑗 + 𝜂𝑡 + 𝛾𝑇𝑟𝑒𝑎𝑡𝑖 ∗ 𝑡 + 𝜀𝑖𝑗𝑡

In equation (2), 𝑌𝑖𝑗𝑡 is a measure of financial aid receipt, re-enrollment, term credits

attempted or earned, or term GPA for student 𝑖 in term 𝑡 at college 𝑗. 𝑇𝑟𝑒𝑎𝑡𝑖 is an indicator for

students who received 5 or more FTE years of Pell. 𝑃𝑜𝑠𝑡2011𝑡 is an indicator for terms on or

after fall 2012 when the new lifetime limit took effect. 𝑃𝑜𝑠𝑡4.5𝐹𝑇𝐸𝑖𝑡 is an indicator set to one in

terms after students attained 4.5 FTE status and is zero otherwise. Because it is unlikely that

treated students far from the eligibility limit in 2012-13 would have immediately changed their

enrollment behavior, this indicator is used to identify when treated students would have been

likely to respond to the policy, if at all, based on proximity to the limit. I define terms post-4.5

FTE as treated terms given that the DOE initially only notified this group of students about the

rule change and later developed warning alerts in the aid processing system specifically for

them.14

𝛼1 captures the effect estimate of interest. It represents the average difference in

outcomes of 5 Pell FTE versus other 5 FTE students in terms after versus before exceeding 4.5

FTE years and after versus before the policy change. In theory, one might expect policy effects

to vary by the size of the Pell Grants students received. Unfortunately, award amounts vary too

little in the matched sample to model effects continuously.15

I therefore estimate average

14

In equation (2), The 𝑃𝑜𝑠𝑡4.5𝐹𝑇𝐸𝑖𝑡 indicator also accounts for secular enrollment trends that arise as students

spend more time in college (e.g. if students naturally tend to re-enroll at higher rates or take larger courseloads as

they get closer to graduation). 15

For example, the 25th

percentile of the eligible award amount in the matched sample is $5,000 (in 2016 dollars)

and the maximum eligible amount is $6,050.

15

treatment effects but examine if effects vary for students eligible for awards less than versus

more than $5,000.

To increase the precision of the estimates and reduce bias, I also include a vector of

individual-level covariates (𝑋𝑖) not used in the matching procedure. This set is comprised of the

following controls: race dummy variables (Black, Hispanic, Asian, Other, and Missing

race/ethnicity), indicators for U.S. citizenship status and Georgia residency status, an indicator of

whether each student initially pursued a bachelor's degree at entry, and an indicator of whether

each student was assigned to remedial coursework at entry, as well as continuous measures of

age at entry, SAT math and verbal scores (imputed where missing), and the student's Expected

Family Contribution at entry (imputed where missing).16

The model also includes school (𝜔𝑗)

and term (𝜂𝑡) fixed effects, as well as a linear term trend (𝛾) that allows enrollment patterns to

vary by treatment status. Finally, assuming that the differences in outcomes between 5 Pell FTE

and other 5 FTE students would not have changed over time in the absence of the eligibility

change, 𝜀𝑖𝑗𝑡 represents a mean-zero random error term. In all estimates, I report standard errors

that account for the potential clustering of schooling behavior within USG campuses.17

Because degree attainment is a singular event for most students, effects on the probability

of bachelor’s degree completion and time to degree cannot be estimated using the same student-

by-term framework as in equation (2). I therefore estimate degree effects at the student level

using a difference-in-differences (DD) design, where the first difference is before versus after the

policy change and the second difference is whether or not a student received 5 FTE years of Pell.

This model takes the following form:

16

Missing SAT scores and EFC at entry are predicted using the full set of non-missing baseline characteristics. In all

results, I present estimates from multiple imputation regressions that account for uncertainty in the imputed values

for students with missing data. 17

I cluster standard errors by the 37 unique campuses prior to USG consolidation activities.

16

(3) 𝑌𝑖𝑗 = 𝛽1𝑇𝑟𝑒𝑎𝑡𝑖 ∗ 𝑃𝑜𝑠𝑡2011𝑖 + 𝛽2𝑃𝑜𝑠𝑡2011𝑖 + 𝛽3𝑇𝑟𝑒𝑎𝑡𝑖 + 𝜙 𝑋𝑖 + 𝜔𝑗 + 𝜀𝑖𝑗𝑡.

In equation (3), 𝑌𝑖𝑗 is a measure of bachelor’s degree attainment overall or within a specific time

interval (e.g., before 6 FTE years) for student 𝑖 at college 𝑗. All other terms are defined as above,

and the coefficient on the interaction term (𝛽1) is the parameter of interest.

IV. RESULTS

Preliminary Graphical Evidence

Comparing the re-enrollment and degree attainment trends of 5 Pell FTE and other 5

FTE students suggests the lifetime rule change caused treated students to increase their effort and

graduate before losing Pell aid. To illustrate this, I plot re-enrollment rates by treatment and

enrollment status relative to the policy change in Figure 1. In the figure, the solid lines show re-

enrollment rates for students enrolled in the post-policy period and the dashed lines show the

same for students who last enrolled prior to the new lifetime limit. 5 Pell FTE students and other

5 FTE students are denoted by white and black circles, respectively. Differences between the

solid and dashed lines before versus after students attained 5 FTE status approximate the DDD

estimates from equation (2). In terms before 5 FTE attainment, treated students re-enrolled at

slightly higher rates than comparison students; however, the relative difference between groups

remained constant over time. Treated students were more likely to re-enroll after receiving 4.5

FTE years of Pell compared to 5 Pell FTE students in the pre-policy period, whereas the re-

enrollment rate among other 5 FTE students remained unchanged before versus after the new

lifetime limit took effect.

Importantly, in each time period 5 Pell FTE and other 5 FTE students made nearly

identical enrollment decisions in the first four years of college. Those results, presented in Figure

2, provide additional evidence that the spike in re-enrollment observed among treated students is

17

unlikely to be a random artifact of the data. Descriptive patterns of time to bachelor’s degree

completion also provide visual evidence of an acceleration effect. In Figure 3, I plot rates of

bachelor’s degree completion before 6 FTE years for 5 Pell FTE (dashed line) and other 5 FTE

(solid line) students by year of 5 FTE attainment. The trends show that both groups of students

followed similar attainment trajectories prior to 2012, yet after the policy change, degree

attainment before reaching the point of aid exhaustion spiked for treated students and exceeded

the rates of other 5 FTE students for the first time.

First-Stage Effects on Financial Aid Receipt

Table 3 shows estimated effects of the rule change on financial aid receipt. In column 1, I

report estimates from a student-by-year panel restricted to four terms preceding and three terms

following 5 FTE attainment, with terms after 4.5 FTE years defined as treated terms. After the

rule change took effect, 5 Pell FTE students received $443 less Pell aid per term on average,

which represents a 33 percent reduction in grant aid. As discussed above, treated students did not

lose their Pell aid in full because students who received 5 FTE years of Pell at the beginning of

an award year remained eligible for full awards. The results in panel B indicate that students did

not replace lost Pell Grant dollars with other sources of grant aid.18

However, they did

compensate by borrowing more – $408 on average per term – which amounts to a 15 percent

increase in loan receipt. The share of aid 5 Pell FTE students received via loans thereby

increased from 66 percent to 76 percent after the new eligibility regime took effect. In column 2

of Table 3, I show that the results are similar if I condition the data on only terms in which

students attended college. In Appendix Table A1, I further show that the estimates are robust to

additional sample restrictions and to the inclusion of student fixed effects, which account for any

18

This in part reflects the fact that Georgia does not offer state need-based aid to students.

18

estimation bias due to time-invariant unobserved student characteristics that are correlated with

aid receipt and exposure to the rule change. In summary, the net effect of the policy change on

total aid dollars appears to be zero, but treated students received more of their aid in the form of

loans after exhausting Pell Grant eligibility.

Effects on Re-enrollment, Credits, and GPA

In Table 4, I report results from estimation of equation (1) on short-term academic

outcomes to examine whether increased borrowing altered students’ enrollment behavior. As in

table 3, I report results using the full student-by-term sample in column 1 and an enrollment-

conditioned sample in column 2. Consistent with the graphical evidence presented earlier, the

coefficient in panel A indicates that after attaining 4.5 FTE status, term-over-term re-enrollment

increased by 2.8 percentage points (3 percent) for treated students. The estimates in panels B and

C indicate the new lifetime limit also increased the number of credits treated students attempted

and earned per term by approximately 0.5 credits.

The enrollment-conditioned estimates in column 2 shed light on whether the overall

impacts in column 1 are driven by effects on the extensive margin (i.e., by increasing re-

enrollment) or on the intensive margin (i.e., by inducing enrolled students to increase their

academic effort). The point estimates in panels B and C of column 2 are roughly one-half the

magnitude of the estimates in column 1, and the coefficient on credits completed remains

statistically significant. This provides some evidence that the lifetime limit affected not only

whether students re-enrolled in college but the intensity with which they did so. In panel D of

column 2, I find no evidence that losing aid affected the grades students earned. The coefficient

on term GPA is small (0.026) and effects larger than .095 points can be ruled out. As in the case

of the estimates on financial aid receipt, the effects on short-term academic outcomes appear

19

robust to alternative sample restrictions and the inclusion of student fixed effects. Results of

these robustness checks are presented in Appendix Table A2.

Effects on Bachelor’s Degree Attainment and Time to Completion

To examine effects on the probability of degree completion and time to degree, I turn to

results from estimation of equation (3). Table 5 presents the main results with and without the

inclusion of student-level controls in even- and odd-numbered columns, respectively. In the first

two columns, the estimates on the interaction term indicate that losing aid did not affect the

overall likelihood that students graduated before exhausting their aid eligibility. The coefficients

with and without controls are both near zero and non-significant. However, this null effect is not

surprising given that more than 80 percent of treated students in the study sample attained 5 Pell

FTE status on or before 2012-13 and therefore had little time to graduate before reaching the

lifetime limit.

In the next four columns, the estimates indicate that the new lifetime rule increased the

probability that treated students graduated in years 6-8 following entry. The coefficient in

column 4 is suggestive of a 2.4 percentage point (3 percent) increase in bachelor’s degree

attainment within 8 years, although it is only marginally significant at the 10 percent level. In

column 6, the coefficient on degree completion in years 6-8 is larger in magnitude and more

precisely estimated. Reducing the lifetime Pell limit increased the probability of bachelor’s

degree completion within this timeframe by 3.1 percentage points, or 7 percent. The last two

columns of Table 5 report effects on bachelor’s degree attainment overall. The coefficients are

positive, but one-third the size of the effects on completion in years 6-8 and not distinguishable

from zero. Taken together, this evidence suggests that treated students were no more likely to

20

graduate overall or before reaching the eligibility limit, but nevertheless accelerated their time to

completion after losing access to Pell aid.

The graphical evidence in Figure 3 suggests that pooled analyses may conceal dynamic

effects on time to completion according to how much lead time students had to react to the new

lifetime limit. In Table 6, I therefore examine degree effects by year of 5 FTE attainment. As

expected, the results in column 1 indicate that treated students who attained 5 FTE status before

or in the first year of the new regime were no more likely to graduate before exhausting their

eligibility for aid. However, students who attained 5 FTE status after 2012-13 had more

opportunity to adjust and were increasingly likely to graduate before their 6 FTE year. The effect

estimates for students who attained 5 FTE status in 2013-14 and 2014-15 are 6.7 percentage

points and 17.5 percentage points, respectively, which represent relative impacts of 15 percent

and 53 percent over comparison group students who attained 5 FTE status in the same years.

I also reject that the impacts on graduating before aid exhaustion are time invariant (the

p-value on an F-test of equal effects is less than .001). By contrast, I find no evidence of dynamic

effects on bachelor’s degree attainment within 8 years or overall in columns 2 and 3 of Table 6.19

Thus, treated students appear to have reacted to the new lifetime limit consistently (i.e., by

accelerating time to completion), but only treated students who maintained more than one year of

eligibility when the new limit took effect increased their likelihood of graduation before

exhausting Pell aid.

I also find evidence that effects on time to completion vary by the amount of Pell Grant

aid students were eligible to receive, although it is unclear a priori whether effects should rise or

19

Although the coefficient on overall degree attainment for students who attained 5 FTE status in 2014-15 is

substantively large (0.044), this is likely an overestimate since most students in this group are members of the 2008

entry cohort and I only observe degree completion through 8 years for this group. Because the point estimates on

overall attainment are near zero for students I observe over a longer time horizon, I expect the estimate for the 2014-

15 group would attenuate over time.

21

fall with financial need. On the one hand, students who are eligible for larger awards stand to

lose more and might increase their effort more as a result. However, if students with greater

financial need also juggle more commitments outside of school, they may have less flexibility to

increase their enrollment intensity. In panel A of Table 7, I disaggregate the effects for students

who were eligible for Pell Grants of less than versus more than $5,000 to examine this question.

All of the acceleration effects are driven by students eligible for less than $5,000 of aid. Degree

completion prior to 6 FTE years increased by 7.6 percentage points for this group, whereas the

point estimate for students eligible for larger awards is slightly negative (-0.019) and not

significant, and I reject that the effects are equal for both groups (the p-value from the joint F-test

is .009).

In panel B of Table 7, I examine variation in effects by total credit completion at the start

of students’ 5 FTE year. Students further from degree completion should be most responsive to

the rule change since students close to degree completion may not need to alter their enrollment

at all to graduate. For this analysis, I also divide students into two mutually exclusive groups:

those who completed fewer than versus greater than 120 credits, with 120 credits serving as a

proxy for proximity to graduation.20

As predicted, all of the acceleration effect is driven by

students who completed fewer than 120 credits at the start of their 5 FTE year. In column 1, the

estimated effect on completion before the point of aid exhaustion for this group is 5 percentage

points. The effect on degree completion within 8 years in column 2 is 6.6 percentage points. The

analogous point estimates for students who completed at least 120 credits at the start of their 5

FTE year are both negative (-0.032 and -0.012, respectively) and not significant. Once again, I

20

Bachelor’s degree programs in the USG system require completion of 120 credits at minimum, but graduates on

average earn 13 excess credits (Complete College America, 2011).

22

reject that effects by total credit attainment are homogenous (the p-values from the joint F-tests

are 0.041 and 0.027, respectively).

Robustness Check 1: Did Concurrent Policy Changes Drive Acceleration Effects?

The new lifetime limit was enacted in the aftermath of the Great Recession when federal

and state governments cut many programs to save costs. In this section, I explore whether the

acceleration effects can be attributed to changes to other financial aid programs made around the

same time. I examine two concurrent changes to financial aid policy which took effect in 2011-

12: 1) elimination of the summer Pell Grant provision, which in 2009-10 and 2010-11 allowed

eligible students to receive more than one Pell Grant in an award year, and 2) changes to

eligibility requirements for Georgia’s merit-aid HOPE Scholarship, which reduced award

amounts for some students already enrolled in college who previously received grants covering

full tuition costs.21

If the effects are partly a function of eligibility changes to the state merit aid

program, then the point estimates should be larger for merit award recipients. Likewise, if the

acceleration effects are due to the year-round Pell and not the new lifetime limit, the effects

should be larger for students who received two Pell awards in the same year.

I present the results of these analyses in Table 8. Panel A reports effects on degree

completion separately for students who did and did not receive HOPE funding in their 5 FTE

year. Panel B reports analogous results for students who did and did not receive two Pell awards

during the 2009-10 or 2010-11 award years. Across all degree outcomes, I fail to reject that

effects for HOPE and non-HOPE recipients and for dual Pell versus non-dual Pell recipients are

equal (none of the p-values from joint F-tests are below 0.25). Furthermore, the magnitudes of

the point estimates are suggestive of larger effects for non-HOPE recipients and for non-dual Pell

21

In 2010-11, approximately one-third of USG students received HOPE Scholarships. On average, award amounts

the following year declined by $300 per semester for students who no longer qualified for full HOPE scholarships

(Suggs, 2016). Details about the summer Pell provision are provided in the text on p.9.

23

recipients. For instance, in panel A of column 1, the estimated effect on completion before

exhausting Pell eligibility is 1.3 percentage points for non-HOPE recipients and -1.5 percentage

points for HOPE recipients. In panel B of column 1, the point estimates for non-dual and dual

Pell recipients are 2.1 and -1.8 percentage points, respectively. The acceleration effects thus do

not appear to be driven by coinciding changes to either the federal Pell Grant or Georgia’s merit-

based aid program.

Robustness Check 2: Stability of Effect Estimates to Alternative Matching Solutions

In addition to concurrent policy changes, another possible concern is that the results may

be sensitive to the choice of covariates used in matching. I therefore examine the stability of the

estimates on degree attainment to alternative estimation and matching solutions in Table 9.

Column 1 reports the main results from Table 5 for purposes of comparison. To examine whether

matching is necessary in the first place, column 2 presents estimates from the main matched

sample, but ignoring the matching solution and instead controlling for the full set of covariates

used in the matching procedure. The results in columns 1 and 2 differ only with respect to the

weights used in estimation. In column 1, the weights depend on the fraction of 5 Pell FTE

students in each matched cell. Strata with more treated students are weighted more heavily to

obtain the estimand of interest: the average treatment effect on the treated. In column 2, strata

with equal shares of 5 Pell FTE and other 5 FTE students receive greatest weight to minimize the

variance of the effect estimate. As a result, the estimates in column 2 will only approximate the

average treatment effect on the treated if effects are homogenous over students. Consistent with

the evidence of heterogeneous effects in Table 7, the estimated effect on degree completion in

years 6-8 in column 2 of Table 9 is 3 times larger than the estimate in column 1 (9.2 versus 3.1

24

percentage points, respectively) because treatment effects vary over strata. Ignoring the matched

solution thus yields upward-biased estimates of the new lifetime limit on time to completion.

Columns 3-5 of Table 9 report estimates of degree effects from three alternative samples

that match 5 Pell FTE to other FTE students using fewer observable characteristics. In column 3,

students were only matched according to their enrollment status post-2011 and cohort of entry.

In column 4, I added the enrollment trajectory measures (i.e., whether students enrolled

continuously in years 1-4, total credits attempted at the start of students’ 5 FTE year, and years to

5 FTE attainment) to the matching procedure. In column 5, gender and race were added. In

column 6, I included EFC status in students’ 5 FTE year but excluded gender and race. The

results indicate that the effect estimates are upward-biased when the enrollment trajectory

measures are excluded from the matching procedure. For example, the point estimate on degree

completion in years 6-8 is 6.3 percentage points in column 3 compared to 2.5 percentage points

in column 4. However, once the enrollment trajectory measures are included, the estimates on

degree completion within 8 years and in years 6-8 increase when imbalance is reduced further by

also matching on sex, race, and EFC. The effects in the fully matched sample in column 1 are

also more precisely estimated than in columns 4-6, and therefore do not suffer from a bias-

variance tradeoff. Taken together, the results in Table 9 suggest that matching is necessary to

obtain unbiased causal estimates and estimates from the fully matched sample capture true policy

effects.

V. CONCLUSION

Despite large public investments in financial aid and concern that too many college

students are not finishing college or taking too long to do so, little is known about whether aid

disbursement policies influence student decisions over whether and when to graduate.

25

Leveraging a recent eligibility change to the lifetime availability of federal Pell Grants, my

findings reveal that students are responsive to lifetime eligibility limits for need-based aid and

that these limits can be designed to accelerate time to completion. Students who exhausted full or

partial eligibility for Pell Grant aid made up for the loss by borrowing 15 percent more on

average per term to pay for college. This in turn affected their subsequent enrollment decisions.

Reducing the lifetime Pell limit from 9 to 6 FTE years increased the probability of term-over-

term re-enrollment and degree completion in years 6-8 since entry increased by 3 percentage

points. Assuming that 400,000 students were initially affected by the new lifetime limit, this

translates into 12,000 students who accelerated their time to completion in response.22

Degree

completion before 6 FTE years also increased 7-18 percentage points (15-53 percent) for

students who had at least one year to adjust to the new lifetime limit before exhausting their aid

eligibility. Importantly, reducing the lifetime availability of Pell aid did not affect the overall

probability of bachelor’s degree completion.

The results from this study therefore indicate that low-income students who have

demonstrated a sustained commitment to earning a degree can be encouraged to graduate more

quickly by setting limits on the availability of need-based aid. An obvious question is whether

the findings would persist if aid eligibility limits affected more students. The findings from this

study should be extrapolated cautiously because the answer to this question is unclear. On the

one hand, Congress restricted Pell Grant eligibility to 6 FTE years on the basis of fiscal

necessity, not because this limit was known to benefit students. It is therefore possible that more

stringent lifetime limits would produce similar acceleration effects. However, students impacted

by the current Pell lifetime limit are also distinct with respect to their commitment to degree

22

Fifty-one percent of 5 Pell FTE students not enrolled post-2011 graduated in years 6-8 versus 54 percent of treated

students enrolled post-2011. As a result, 12,000 students [(400,000*0.54)-(400,000*.51)] are estimated to have

graduated more quickly. See footnote 8 for how the estimated number of affected students is derived.

26

completion. At some point, excessively stringent aid limits would certainly introduce costs (e.g.,

increased dropout) that outweigh the benefits realized from the current policy. The tipping point

at which lifetime aid limits begin to do more harm than good remains an important question left

for further research.

A related question is whether reallocating how aid is disbursed along the path to

completion would produce greater benefits than simply eliminating aid that is disbursed late into

college. Several recent studies have found that disbursing more generous need-based aid to

students in the first two years of college can increase postsecondary attainment and generate

positive returns on investment (Angrist et al., 2016; Castleman & Long, 2016; Goldrick-Rab et

al., 2016). It may therefore prove more cost-effective to maintain spending on need-based aid but

frontload those dollars to help students establish their footing in college.23

Finally, this study arrives at a time when free college plans are widely popular.

Legislation making college tuition-free for certain groups of students has passed in six states as

of 2016, and 17 additional states are actively considering legislation (Pingel, Parker, & Sisneros,

2016). Place-based scholarships have also become a popular strategy for cities to reduce local

college costs and retain its young talent. Despite their intention of making college more

affordable, free college plans may have adverse effects, such as incentivizing attendance at

institutions where students are less likely to graduate (Svrluga, 2015). The results in this study

offer empirical evidence that efforts to reduce time to completion may be most effective when

students and families have some skin in the game.

23

I estimate that Pell awards to the approximately 2.1 million first-year recipients per year could be increased by

$705 per year on average if the cost savings from the new limit ($1.48 billion if 400,000 students lost eligibility)

were redistributed.

27

REFERENCES

Alsalam, N. (2013). The Federal Pell Grant Program: Recent Growth and Policy Options.

Washington, DC: Congressional Budget Office.

Angrist, J., Autor, D., Hudson, S., & Pallais, A. (2016). Evaluating post-secondary aid:

Enrollment, persistence, and projected completion effects (NBER Working Paper No.

23015). Cambridge: National Bureau of Economic Research.

Baum, S., Elliott, D. C., & Ma, J. (2014). Trends in Student Aid 2014. Washington, DC:

CollegeBoard.

Bettinger, E., Gurantz, O., Kawano, L., & Sacerdote, B. (2016). The long run impacts of merit

aid: Evidence from California’s Cal Grant (NBER Working Paper No. 22347). Cambridge:

National Bureau of Economic Research.

Bettinger, E. P., Boatman, A., & Long, B. T. (2013). Student Supports: Developmental

Education and Other Academic Programs. The Future of Children, 23(1), 93–115.

Boatman, A., Evans, B. J., & Soliz, A. (2017). Understanding Loan Aversion in Education.

AERA Open, 3(1), 1–16.

Bound, J., Lovenheim, M. F., & Turner, S. (2012). Increasing Time to Baccalaureate Degree in

the United States. Education Finance and Policy, 7(4), 375–424.

Carruthers, C. K., & Özek, U. (2016). Losing HOPE: Financial aid and the line between college

and work. Economics of Education Review, 53, 1–15.

Castleman, B. L., & Long, B. T. (2016). Looking beyond enrollment: The causal effect of need-

based grants on college access, persistence, and graduation. Journal of Labor Economics,

34(4), 1023–1073.

Complete College America (2001). Time is the enemy. Washington, DC: Complete College

America.

Conger, D., & Turner, L. J. (2015). The impact of tuition increases on undocumented college

students’ attainment (NBER Working Paper No. 21135). Cambridge: National Bureau of

Economic Research.

Cornwell, C. M., Lee, K. H., & Mustard, D. B. (2005). Student Responses to Merit Scholarship

Retention Rules. The Journal of Human Resources, 40(4), 895–917.

Delisle, J., & Miller, B. (2015). Myths & misunderstandings: The undeserved legacy of year-

round Pell Grants. Washington, DC: New America.

DesJardins, S. L., Ahlburg, D. A., & McCall, B. P. (2002). Simulating the longitudinal effects of

changes in financial aid on student departure from college. The Journal of Human

Resources, 37(3), 653–679.

Dynarski, S. M. (2003). Does aid matter? Measuring the effect of student aid on college

attendance and completion. American Economic Review, 93(1), 279–288.

Dynarski, S., & Scott-Clayton, J. (2013). Financial Aid Policy: Lessons from Research. Future of

Children, 23(1), 67–91.

28

Gichevu, D., Ionescu, F., & Simpson, N. (2012). The effects of credit status on college

attainment and college completion (No. IZA Discussion Paper No. 6719). Bonn, Germany:

Institute of Labor Economics.

Goldrick-Rab, S., & Kelchen, R. (2015). Making sense of loan aversion: Evidence from

Wisconsin. In B. Hershbein & K. M. Hollenbeck (Eds.), Student Loans and the Dynamics of

Debt (pp. 317–378). Kalamazoo, MI: W.E. Upjohn Institute for Employment Research.

Goldrick-Rab, S., Kelchen, R., Harris, D. N., & Benson, J. (2016). Reducing income inequality

in educational attainment: Experimental evidence on the impact of financial aid on college

completion. American Journal of Sociology, 121(6), 1762–1817.

Iacus, S. M., King, G., & Porro, G. (2012). Causal inference without balance checking:

Coarsened exact matching. Political Analysis, 20(1), 1–24.

Kahneman, D., & Tversky, A. (1979). Prospect Theory: An Analysis of Decision under Risk

Econometrica, 47(2), 263–292.

King, G., Nielsen, R., Coberley, C., & Pope, J. E. (2011). Comparative effectiveness of matching

methods for causal inference. Cambridge: Harvard University. Unpublished manuscript.

Mahan, S. M. (2011). Federal Pell Grant Program of the Higher Education Act: Background,

Recent Changes, and Current Legislative Issues. Washington, DC: Congressional Research

Service.

Nelson, L. (2012, October 31). Some college, no degree, no Pell. Inside Higher Ed.

Pingel, S., Parker, E., & Sisneros, L. (2016). Free community college: An approach to increase

adult student success in postsecondary education. Denver: Education Commission of the

States.

Richburg-Hayes, L., Brock, T., LeBlanc, A., Paxson, C., Rouse, C. E., & Barrow, L. (2009).

Rewarding persistence: Effects of a performance-based scholarship program for low-

income parents. New York: MDRC.

Schudde, L., & Scott-Clayton, J. (2016). Pell Grants as performance-based scholarships? An

examination of Satisfactory Academic Progress requirements in the nation’s largest need-

based aid program. Research in Higher Education, 57(8), 943–967.

Scott-Clayton, J. (2011). On Money and Motivation: A Quasi-Experimental Analysis of

Financial Incentives for College Achievement. Journal of Human Resources, 46(3), 614–

Scott-Clayton, J. (2012). What Explains Trends in Labor Supply Among U.S. Undergraduates?

National Tax Journal, 65(1), 181–210.

Suggs, C. (2016). Troubling gaps in HOPE point to need-based aid solutions. Atlanta: Georgia

Budget and Policy Institute.

Svrluga, S. (2015, January 14). Unintended consequences of President Obama’s proposal to

make community college free, and one key chart. The Washington Post.

U.S. Department of Education (2012). Implementation of the 12 Semester Lifetime Limit for

Federal Pell Grants. Retrieved August 9, 2017, from

https://ifap.ed.gov/eannouncements/021712PreliminaryInfoImpl12SemesterFedPellGrant.ht

ml.

29

Table 1. Summary statistics of full 5 FTE year sample and matched sample by treatment status

(1) (2) (3) (4) (5) (6)

Matched Sample

All students enrolled

for 5+ FTE years

Treated: Enrolled for 5+

FTE years and received

5+ years of Pell

Comparison: Enrolled

for 5+ FTE years and

received < 5 years of

Pell

Mean SD Mean SD Mean SD

Pre-entry characteristics

Female 0.561 0.496 0.624 0.484 0.624 0.484

Black 0.405 0.491 0.613 0.487 0.613 0.487

Asian 0.063 0.243 0.063 0.244 0.060 0.237

Latino 0.045 0.208 0.039 0.193 0.042 0.201

White 0.447 0.497 0.251 0.434 0.251 0.434

Race Other 0.029 0.167 0.024 0.153 0.023 0.150

Missing Race 0.011 0.103 0.010 0.101 0.011 0.104

U.S. Citizen 0.942 0.235 0.936 0.245 0.932 0.252

GA Resident 0.970 0.171 0.976 0.154 0.961 0.195

BA Degree Program at Entry 0.978 0.148 0.970 0.170 0.978 0.147

SAT Math + Verbal Score 1012 159 928 146 950 147

Missing SAT 0.05 0.217 0.064 0.245 0.046 0.21

Assigned to Remedial Coursework 0.189 0.392 0.244 0.430 0.194 0.395

Age at Entry 18.63 0.91 18.72 1.27 18.69 1.03

EFC at Entry $14,434 $20,344 $1,092 $2,975 $6,642 $6,702

Missing EFC at Entry 0.155 0.361 0.014 0.119 0.206 0.405

Post-entry characteristics

Age 5 FTE Year 23.55 1.27 23.51 1.56 23.52 1.31

EFC in 5 FTE Year $8,201 $14,735 $476 $1,029 $534 $1,244

Cum Credits Attempted at Start of 5 FTE Year 135.30 13.58 136.20 15.74 136.28 13.32

Cum Credits Earned at Start of 5 FTE Year 118.83 15.61 120.46 15.92 120.78 15.09

Cum GPA at Start of 5 FTE Year 2.70 0.52 2.64 0.50 2.65 0.52

Terms to 5 FTE Status 15.71 2.47 15.34 2.33 15.42 2.33

Total Pell FTE Years 2.07 2.26 5.57 0.50 3.20 1.35

Observations 46,766 8,656 7,932

Source: University System of Georgia administrative records.

30

Table 2. Matched sample characteristics of treated and comparison students by enrollment status relative to the Pell eligibility rule change

(1) (2) (3) (4) (5) (6) (7)

Treated Students Comparison Students

Did Not

Enroll

After Fall

2011

Enrolled

On or

After

Spring

2012 Post - Pre

Did Not

Enroll

After Fall

2011

Enrolled

On or

After

Spring

2012 Post - Pre DID

Pre-entry characteristics

Female 0.627 0.622 -0.005 0.627 0.622 -0.005 0.000

(0.008) (0.023) (0.024)

Black 0.631 0.600 -0.030 0.631 0.600 -0.030 -0.000

(0.018) (0.022) (0.023)

Asian 0.063 0.063 0.000 0.055 0.063 0.008* -0.008

(0.006) (0.005) (0.006)

Latino 0.020 0.052 0.032*** 0.027 0.053 0.025*** 0.007

(0.005) (0.006) (0.008)

White 0.262 0.243 -0.018 0.262 0.243 -0.018 0.000

(0.018) (0.019) (0.020)

Race Other 0.021 0.026 0.005 0.020 0.025 0.005 -0.000

(0.003) (0.005) (0.005)

Missing Race 0.003 0.015 0.012*** 0.005 0.015 0.010*** 0.001

(0.002) (0.004) (0.003)

SAT Math + Verbal Score 943 917 -26.232*** 958 945 -12.665* -13.568

(5.264) (6.772) (8.249)

Assigned to Remedial Coursework 0.219 0.262 0.042*** 0.180 0.203 0.023 0.019

(0.015) (0.019) (0.021)

Age at Entry 18.72 18.71 -0.009 18.67 18.71 0.047 -0.056

(0.030) (0.035) (0.045)

EFC at Entry $967 $1,178 206.392*** $6,349 $6,843 539.191* -332.798

(70.486) (268.983) (280.382)

Post-entry characteristics

Age 5 FTE Year 23.50 23.52 0.016 23.46 23.55 0.090** -0.074

31

(0.046) (0.044) (0.051)

EFC in 5 FTE Year $500 $460 -39.207 $565 $513 -52.932* 13.725

(25.528) (30.448) (36.492)

Cum Credits Attempted at Start of 5 FTE Year 137.50 135.30 -2.195*** 137.87 135.18 -2.681*** 0.486

(0.730) (0.659) (0.743)

Cum Credits Earned at Start of 5 FTE Year 119.47 121.15 1.688*** 120.11 121.23 1.122** 0.566

(0.559) (0.543) (0.500)

Cum GPA at Start of 5 FTE Year 2.67 2.62 -0.046*** 2.66 2.64 -0.017 -0.029

(0.015) (0.025) (0.023)

Observations 3,532 5,124 8,656 3,446 4,486 7,932 16,588

*** p<0.01 ** p<0.05 * p<0.10

Notes: Means are reported in columns (1), (2), (4), and (5). Estimates of pre-post compositional differences are reported in columns (3) and (6). Estimates of the

difference in pre-post differences between treated and comparison students are reported in column (7). Standard errors are clustered by institution and reported in

parentheses.


32

Table 3. Estimates of the effect of the Pell Grant eligibility change

on financial aid receipt in terms following 5 FTE attainment

(1) (2)

All student-

by-term

observations

Restricted to

enrolled

terms only

A. Pell Grant Aid -443.288*** -582.163***

(58.116) (57.032)

R2 0.220 0.202

Baseline mean $1,327 $1,826

B. Other grant aid 30.150 49.772

(25.108) (32.955)

R2 0.092 0.098

Baseline mean $53 $74

C. Loans 408.054*** 454.952**

(145.994) (195.708)

R2 0.131 0.158


D. Total financial aid 7.540 -60.062

(166.901) (210.731)

R2 0.180 0.196


Student-by-term observations 104,596 92,545

*** p<0.01 ** p<0.05 * p<0.10

Note: The analytic window is restricted to four terms preceding and

three terms following 5 FTE attainment, with terms after 5 FTE years

defined as treated terms. Student-by-term observations following

bachelor's degree receipt and summer terms are excluded. Results are

estimated with least squares models that include the full set of controls.

See table 5 for details. All models also include linear time trends allowed

to vary by treatment status and before versus after the policy change.

Robust standard errors, clustered by institution, are shown in

parentheses.


33

Table 4. Estimates of the effect of the Pell Grant eligibility change

on the probability of re-enrollment, credits attempted and earned,

and GPA in terms following 5 FTE attainment

(1) (2)

All student-

by-term

observations

Restricted to

enrolled

terms only

A. Re-enrolled 0.028***

(0.010)

R2 0.154

Baseline mean 0.835

B. Term credits attempted 0.473** 0.225

(0.184) (0.140)

R2 0.161 0.070

Baseline mean 10.39 12.41

C. Term credits earned 0.470*** 0.295**

(0.163) (0.144)

R2 0.158 0.073

Baseline mean 9.13 10.90

D. Term GPA 0.026

(0.035)

R2 0.080

Baseline mean 2.55

Student-by-term observations 104,596 92,545

*** p<0.01 ** p<0.05 * p<0.10

Note: The analytic window is restricted to four terms preceding and

three terms following 5 FTE attainment, with terms on or after 5 FTE

years defined as treated terms. Student-by-term observations following

bachelor's degree receipt and summer terms are excluded. Results are

estimated with least squares models that include the full set of controls.

See table 5 for details. All models also include linear time trends allowed

to vary by treatment status and before versus after the policy change.

Robust standard errors, clustered by institution, are shown in

parentheses.


34

Table 5. Estimates of the effect of the Pell Grant eligibility change on bachelor's degree attainment overall and time to degree completion

(1) (2) (3) (4) (5) (6) (7) (8)

BA before

6 FTE year

BA w/in

8 years of entry

BA in

year 6-8 since entry

BA

ever

Post-2011 x Treated 0.004 0.006 0.025* 0.024* 0.036** 0.031** 0.012 0.010

(0.017) (0.018) (0.014) (0.013) (0.016) (0.014) (0.016) (0.016)

Treated -0.031** -0.030** -0.002 0.006 0.030** 0.054*** 0.006 0.020*

(0.013) (0.013) (0.014) (0.012) (0.014) (0.013) (0.012) (0.011)

Post-2011 -0.169*** -0.159*** -0.096*** -0.085*** 0.001 -0.003 -0.090*** -0.081***

(0.017) (0.016) (0.015) (0.016) (0.017) (0.016) (0.017) (0.018)

R2 0.029 0.088 0.010 0.091 0.003 0.047 0.011 0.090

Controls N Y N Y N Y N Y

Comparison mean post-2011 0.544 0.728 0.454 0.769

Observations 16,588

*** p<0.01 ** p<0.05 * p<0.10

Note: Results are estimated with linear probability models. Models with controls are from multiple imputation specifications that include the following

covariates: race dummy variables (Black, Hispanic, Asian, Other, and Missing race/ethnicity); U.S. citizen dummy variable; Georgia resident dummy

variable; pursued bachelor's degree at entry dummy variable; assigned to remedial coursework at entry dummy variable; SAT math and verbal scores

(imputed where missing); age at entry; and the student's Expected Family Contribution at entry (imputed where missing). All models with controls also

include institution fixed effects and a constant. Standard errors are clustered by institution and reported in parentheses.


35

Table 6. Estimates of the effect of the Pell Grant eligibility change on bachelor's degree

attainment by year of 5 FTE attainment

(1) (2) (3)

Completion

outcomes

BA before

6 FTE year

BA w/in

8 years of

entry

BA

ever

5 FTE in 2012-13 or earlier -0.013 0.020 0.007

(0.019) (0.015) (0.018)

5 FTE in 2013-14 0.067* 0.032 0.006

(0.038) (0.036) (0.031)

5 FTE in 2014-15 0.175*** 0.038 0.044

(0.052) (0.054) (0.064)

Tests of equal effects (p-values) <.001 0.357 0.494

R2 0.092 0.105 0.108

Comparison mean in 2012-13 or earlier 0.569 0.760 0.801

Comparison mean in 2013-14 0.462 0.639 0.683

Comparison mean in 2014-15 0.328 0.415 0.442

Observations 16,588

*** p<0.01 ** p<0.05 * p<0.10 Note: Results are estimated with linear probability models that include the full set of controls.

See table 5 for details. Standard errors are clustered by institution and reported in parentheses.


36

Table 7. Estimates of the effect of the Pell Grant eligibility change on bachelor's degree attainment

by eligible Pell award amount and cumulative credit attainment at start of 5 FTE year

(1) (2) (3)

BA before

6 FTE

year

BA w/in

8 years of

entry

BA

ever

A. Eligible annual Pell award amount (in 2016 dollars)

Less than $5,000 0.076** 0.061** 0.043

(0.028) (0.024) (0.026)

$5,000 to $6,050 -0.019 0.008 -0.004

(0.020) (0.017) (0.020)

Tests of equal effects (p-values) 0.009 0.119 0.165

R2 0.634 0.795 0.83

Comparison mean post-2011: Pell award < $5,000 0.611 0.782 0.813

Comparison mean post-2011: Pell award of $5,000 to

$6,050 0.529 0.715 0.758

Observations 16,588

B. Credits completed at start of 5 FTE term

Less than 120 credits 0.050* 0.066*** 0.035

(0.027) (0.023) (0.026)

120 credits or more -0.032 -0.012 -0.012

(0.021) (0.018) (0.019)


R2 0.668 0.804 0.836

Comparison mean post-2011: Completed < 120 credits 0.349 0.596 0.663

Comparison mean post-2011: Completed 120 credits or

more 0.701 0.834 0.854

Observations 16,588

*** p<0.01 ** p<0.05 * p<0.10

Note: Results are estimated with linear probability models that include the full set of controls. See table 5 for

details. Standard errors are clustered by institution and reported in parentheses.


37

Table 8. Estimates of the effect of the Pell Grant eligibility change on bachelor's degree attainment by

HOPE scholarship receipt in 5 FTE year and receipt of two Pell awards in 2009-10/2010-11

(1) (2) (3)

BA before

6 FTE

year

BA w/in

8 years of

entry

BA

ever

A. Received Georgia HOPE Scholarship in 5 FTE year

No 0.013 0.031** 0.014

(0.021) (0.015) (0.017)

Yes -0.015 -0.002 0.002

(0.037) (0.025) (0.023)


R2 0.638 0.797 0.831

Comparison mean post-2011: Did not receive HOPE 0.509 0.702 0.747

Comparison mean post-2011: Received HOPE 0.770 0.897 0.909

Observations 16,588

B. Received two Pell awards in 2009-10/2010-11

No 0.021 0.024 0.010

(0.025) (0.018) (0.021)

Yes -0.018 0.019 0.006

(0.030) (0.023) (0.026)


R2 0.633 0.795 0.829

Comparison mean post-2011: Did not receive two Pell

awards 0.539 0.704 0.753

Comparison mean post-2011: Received two Pell awards 0.550 0.764 0.792

Observations 16,588

*** p<0.01 ** p<0.05 * p<0.10 Note: Results are estimated with linear probability models that include the full set of controls. See table 5 for

details. Standard errors are clustered by institution and reported in parentheses.


38

Table 9. Robustness of estimates of effects on bachelor's degree attainment overall and time to degree completion to alternative matching solutions

(1) (2) (3) (4) (5) (6)

Main Sample Alternative Matched Samples

Matched Unmatched One Two Three Four

A. BA before 6 FTE year 0.006 -0.011 -0.005 0.004 0.005 0.018

(0.018) (0.015) (0.011) (0.013) (0.016) (0.015)

B. BA w/in 8 years of entry 0.024* 0.019* 0.004 0.008 0.012 0.022

(0.013) (0.011) (0.013) (0.015) (0.016) (0.015)

C. BA in years 6-8 since entry 0.031** 0.092*** 0.063*** 0.025 0.025 0.031

(0.014) (0.013) (0.015) (0.018) (0.019) (0.018)

D. BA ever 0.010 0.009 -0.006 -0.001 -0.010 0.015

(0.016) (0.012) (0.013) (0.015) (0.014) (0.015)

Covariates used in matching solution

Cohort

Post-2011 Enrollment Status

Enrolled Continuously in Years 1-4

Credits Attempted at Start of 5 FTE Year

Years to Attain 5 FTE Status

Gender

Race

EFC in 5 FTE Year

Observations 16,588 26,031 25,995 24,844 20,148

*** p<0.01 ** p<0.05 * p<0.10

Note: Results are estimated with linear probability models that include the full set of controls. See table 5 for details. Results in column 2 also control for the full

set of covariates used in matching instead of weighting by the matched stratum as in column 1. Standard errors are clustered by institution and reported in

parentheses.


39

Figure 1. Re-enrollment by term within three years of the new lifetime Pell limit, by treatment

and enrollment status relative to the policy change

.84

.88

.92

.96

1

Re-

enro

llm

ent

Rat

e

3 3.5 4 4.5 5 5.5 6

FTE Year

Treated, Enrolled Pre-2012 Treated, Enrolled Post-2012

Comparison, Enrolled Pre-2012 Comparison, Enrolled Post-2012

40

Figure 2. Enrollment and credit attainment in years 1-4 of college, by treatment and enrollment

status relative to the policy change

A. Enrollment by term

B. Total credits attempted by term

.9.9

2.9

4.9

6.9

8

1

En

roll

men

t R

ate

1 2 3 4 5 6 7 8

Term

Treat, Enrolled Pre-2012 Treat, Enrolled Post-2012


02

04

06

08

01

00

Cu

mu

lati

ve

Cre

dit

s E

arned

1 2 3 4 5 6 7 8

Term

Treat, Enrolled Pre-2012 Treat, Enrolled Post-2012


41

Figure 3. Bachelor’s degree completion rates before the attainment of 6 FTE status, by treatment

status and year of 5 FTE attainment

Note: Reference line denotes the first year in which the 6 FTE year lifetime Pell limit took effect.

.6.7

.8.9

1

BA

Bef

ore

6 F

TE

Yea

r

2006 2007 2008 2009 2010 2011 2012 2013 2014

Year of 5 FTE Attainment

Treated Comparison

42

Table A1. Robustness of estimates of the effect of the Pell Grant eligibility change on financial aid receipt in terms following 5 FTE attainment

(1) (2) (3) (4) (5) (6) (7) (8)

Not conditioned on enrollment Restricted to enrolled terms only

Baseline

model and

sample

+ Student

fixed effects

Sample

restricted to

balanced

panel

Sample

restricted to

graduates

Baseline

model and

sample

+ Student

fixed effects

Sample

restricted to

balanced

panel

Sample

restricted to

graduates

A. Pell Grant Aid -443.288*** -412.163*** -491.647*** -498.142*** -582.163*** -524.798*** -720.492*** -630.904***

(58.116) (55.273) (85.981) (54.962) (57.032) (55.361) (96.253) (57.253)

R2 0.220 0.521 0.279 0.188 0.202 0.563 0.251 0.200

Baseline mean $1,327 $1,327 $1,020 $1,822 $1,826 $1,826 $1,751 $2,019

B. Other grant aid 30.150 29.744 69.569* 35.896 49.772 33.844 73.787 42.139

(25.108) (36.514) (38.657) (31.685) (32.955) (46.925) (50.875) (40.983)

R2 0.092 0.663 0.076 0.095 0.098 0.700 0.084 0.099

Baseline mean $53 $53 $22 $81 $74 $74 $35 $88

C. Loans 408.054*** 376.561** 616.808** 495.069** 454.952** 328.274* 766.637** 324.885

(145.994) (183.285) (286.449) (187.356) (195.708) (181.517) (319.471) (202.715)

R2 0.131 0.571 0.134 0.151 0.158 0.665 0.172 0.170

Baseline mean $2,714 $2,714 $2,194 $3,381 $3,769 $3,769 $3,787 $3,751

D. Total financial aid 7.540 18.613 218.855 42.834 -60.062 -132.788 153.354 -255.511

(166.901) (216.315) (331.253) (203.457) (210.731) (230.819) (357.189) (215.072)

R2 0.180 0.541 0.183 0.185 0.196 0.641 0.185 0.211

Baseline mean $4,103 $4,103 $3,239 $5,292 $5,679 $5,679 $5,577 $5,868

Student-by-term observations 104,596 104,596 33,512 80,398 92,545 94,731 27,402 76,427

*** p<0.01 ** p<0.05 * p<0.10 Note: The analytic window is restricted to four terms preceding and two terms following 5 FTE attainment, with terms on or after 5 FTE years defined as treated terms. Student-

by-term observations following bachelor's degree receipt and summer terms are excluded. Results are estimated with least squares models that include the full set of controls. See

Table 5 for details. All models also include linear time trends allowed to vary by treatment status and before versus after the policy change. In columns (3) and (7), the sample is

restricted to students who did not graduate before two terms following 5 FTE attainment. Robust standard errors, clustered by institution, are shown in parentheses.


43

Table A2. Robustness of effect estimates on the probability of re-enrollment, credits attempted and earned, and GPA in terms following 5 FTE attainment

(1) (2) (3) (4) (5) (6) (7) (8)

Not conditioned on enrollment Restricted to enrolled terms only

Baseline

model and

sample

+ Student

fixed effects

Sample

restricted to

balanced

panel

Sample

restricted to

graduates

Baseline

model and

sample

+ Student

fixed effects

Sample

restricted to

balanced

panel

Sample

restricted to

graduates

A. Re-enrolled 0.028*** 0.026** 0.056** 0.027***

(0.010) (0.011) (0.023) (0.008)

R2 0.154 0.371 0.212 0.0315

Baseline mean 0.835 0.835 0.704 0.947

B. Term credits attempted 0.473** 0.413* 0.636** 0.632*** 0.225 0.165 -0.047 0.306*

(0.184) (0.207) (0.308) (0.210) (0.140) (0.162) (0.228) (0.167)

R2 0.161 0.393 0.202 0.0701 0.0701 0.350 0.0714 0.0637

Baseline mean 10.39 10.39 8.616 11.98 12.41 12.41 12.25 12.64

C. Term credits earned 0.470*** 0.348* 0.481 0.678*** 0.295** 0.110 0.124 0.388**

(0.163) (0.186) (0.311) (0.187) (0.144) (0.154) (0.300) (0.152)

R2 0.158 0.411 0.200 0.0677 0.0734 0.362 0.0857 0.0567

Baseline mean 9.13 9.134 6.947 11.07 10.9 10.90 9.802 11.69

D. Term GPA 0.026 -0.000 0.098* 0.035

(0.035) (0.038) (0.056) (0.036)

R2 0.0804 0.595 0.0779 0.0803

Baseline mean 2.545 2.545 2.112 2.752

Student-by-term observations 104,596 104,596 33,512 80,398 94,731 94,731 27,402 76,427

*** p<0.01 ** p<0.05 * p<0.10 Note: The analytic window is restricted to four terms preceding and three terms following 5 FTE attainment, with terms on or after 5 FTE years defined as treated terms. Student-

by-term observations following bachelor's degree receipt and summer terms are excluded. Results are estimated with least squares models that include the full set of controls. See

Table 5 for details. All models also include linear time trends allowed to vary by treatment status and before versus after the policy change. In columns (3) and (7), the sample is

restricted to students who did not graduate before two terms following 5 FTE attainment. Robust standard errors, clustered by institution, are shown in parentheses.

Bachelor’s Degree A - Home | Scholars at Harvard · Bachelor’s Degree Attainment and Time to...

Documents

Transcript of Bachelor’s Degree A - Home | Scholars at Harvard · Bachelor’s Degree Attainment and Time to...